Skip to content

The lab

The hardware behind everything we run.

This is the kit everything else on the site runs on: one rack, a GPU node and a four-GPU AI server. Walk it in 3D, or read every spec in the list.

01 Walkthrough

The rack, and what stands beside it

Every unit is modelled in the position it occupies in the real cabinet. Pick a device in the 3D view, or read its card below.

A rendering of the lab: the 42U rack — cooling, patch panel, UniFi switch and gateway, a display, routing and carrier gear, the two HPE ProLiant servers, a shelf, the SFP switch, the MD1220 flash shelf, the GPU node, the R730, power and a UPS — and beside it a desk carrying the open-frame AI server with four cards.

AC Infinity rack exhaust

Pulls heat out of the top of the rack.

Type
2U exhaust fan panel with controller

Cat 6 patch panel

Structured cabling termination.

Ports
24× Cat 6

UniFi Switch Pro Max 24

Core switching for the lab.

Ports
24 switched
Telemetry
Per-port traffic, exported

UniFi Dream Machine Pro SE

Gateway, firewall and VPN hub for everything on this page.

Role
Routing, firewall, DHCP, VPN
Ports
8× RJ45 + SFP+
Telemetry
Exported to the monitoring stack

Rack display

Runs the live network dashboard: the same galaxy view as the wall screen.

Type
ELEMENT monitor on a rack mount

1U router

Modular router under the display.

Type
1U, modular interface slots

Cisco Catalyst, 48-port

48-port access switching in the rack.

Ports
48× RJ45
Uplinks
4× SFP

ADVA FSP 150CC

Carrier Ethernet demarcation: where the provider's circuit ends and the lab begins.

Type
Carrier Ethernet NID
Ports
6× RJ45 + SFP

HPE ProLiant DL560 Gen10

Four-socket compute: the highest core count in the lab once it is running.

CPU
4× Xeon Gold 6152, 2.1 GHz
Cores / threads
88 / 176 (22 per socket)
Memory
256 GB DDR4 ECC
Storage
8 × 1.8 TB SAS (14.4 TB raw)
Controller
HPE Smart Array P408i-a
Bays
8 SFF

HPE ProLiant DL360 Gen10

Dual-socket 1U compute.

CPU
2× Xeon Silver 4208, 2.1 GHz
Cores / threads
16 / 32 (8 per socket)
Memory
128 GB
Storage
4 × 1.2 TB + 3 × 300 GB HDD

Console shelf

Keyboard for hands-on work at the rack.

Type
Cantilever shelf

Cisco ME 3800X

Carrier Ethernet aggregation, with a compact ADVA demarcation unit on the shelf above it.

Ports
24× SFP + 4× RJ45
Uplinks
2× 10G SFP+

Dell PowerVault MD1220

The flash shelf: the bulk of the lab's storage, attached to the R730.

Bays
24 × 2.5"
Drives
24 × 1.8 TB SAS SSD
Raw flash
43.2 TB

GPU node

AI inference, camera analytics and media transcoding.

CPU
AMD Ryzen 9 5950X
Cores / threads
16 / 32
Board
ASUS ROG Crosshair VIII Formula
Memory
64 GB (4 × 16 GB)
GPU
RTX 3080 Ti 12 GB + GTX 1080 Ti 11 GB
VRAM
23 GB
Storage
2 TB NVMe
Guests
27 (22 containers, 5 VMs)

Dell PowerEdge R730

The virtualisation host. Nearly everything runs here.

CPU
2× Xeon E5-2620 v3, 2.4 GHz
Cores / threads
12 / 24
Memory
384 GB (24 × 16 GB, every slot)
Internal disk
7 × 1.2 TB + 1 × 300 GB SAS
Boot / cache
1 TB NVMe
External
The MD1220 flash shelf above it
Usable storage
48.1 TiB ZFS (shelf + internal)
Network
2× 10 GbE + 2× 1 GbE
Guests
86 (68 containers, 18 VMs)

Switched power strip

Ten individually switched outlets.

Outlets
10, switched

AC Infinity rack cooling

Front cooling unit with a temperature display.

Type
2U fan unit with thermostat

APC UPS

Battery backup for the rack.

Type
2U rack-mount UPS

4× RTX 4090 AI server

Local LLM inference and agent workloads. Open-frame, on the desk beside the rack.

CPU
AMD Threadripper
Memory
128 GB DDR4-3200
GPU
4× RTX 4090 24 GB
VRAM
96 GB

02 In service

The totals, with the arithmetic shown

Everything in the rack and on the desk, added up.

Server CPU cores
132
Server threads
264
Memory
960 GB
GPU memory
119 GB
Storage
48.1 TiB usable ZFS + 20.1 TB in the HPE servers
Guests
113

No address, hostname or software version appears here. This is a picture of the hardware, not a map of the estate.